Risky event classification leveraging transfer learning for very limited datasets in optical networks

Khouloud Abdelli; Matteo Lonardi; Jurgen Gripp; Samuel Olsson; Fabien Boitier; Patricia Layec

doi:10.1364/JOCN.517529

Journal of Optical Communications and Networking
Vol. 16,
Issue 7,
pp. C51-C68
(2024)
•https://doi.org/10.1364/JOCN.517529

Risky event classification leveraging transfer learning for very limited datasets in optical networks

Khouloud Abdelli, Matteo Lonardi, Jurgen Gripp, Samuel Olsson, Fabien Boitier, and Patricia Layec

Not Accessible

Your library or personal account may give you access

Get PDF
Email
Share
Get Citation
Copy Citation Text
Khouloud Abdelli, Matteo Lonardi, Jurgen Gripp, Samuel Olsson, Fabien Boitier, and Patricia Layec, "Risky event classification leveraging transfer learning for very limited datasets in optical networks," J. Opt. Commun. Netw. 16, C51-C68 (2024)

Export Citation
- BibTex
- Endnote (RIS)
- HTML
- Plain Text
Citation alert
Save article

Check for updates

Related Topics
Optics & Photonics Topics
?

The topics in this list come from the Optics and Photonics Topics applied to this article.

About this Article
History
- Original Manuscript: January 2, 2024
- Revised Manuscript: March 21, 2024
- Manuscript Accepted: March 21, 2024
- Published: May 6, 2024
Virtual Issues
Journal of Optical Communications and Networking ECOC 2023 (2024)

Abstract

Monitoring the state of polarization (SOP) is crucial for tracking vibrations or disturbances in the vicinity of optical fibers, such as precursors to fiber cuts. While SOP data are valuable for machine learning (ML) models in identifying vibrations, acquiring a sufficient amount of data presents a significant challenge. To overcome this hurdle, we introduce an innovative transfer learning framework designed for the identification of vibrations (events) when confronted with limited SOP data. Our methodology leverages the pre-trained convolutional neural network MobileNet as a feature extractor, incorporating the encoding of time series SOP measurements into images for MobileNet input. We explore different time series encoding techniques, including the Gramian Angular Difference Field (GADF) and the Gramian Angular Summation Field (GASF). Different architectures for building our transfer learning framework based on MobileNet are investigated. Validation of our proposed approaches is conducted using experimental data that simulates movements indicative of fiber break precursors. The experimental results clearly demonstrate the superior performance of our approaches compared to other ML algorithms, especially in scenarios with limited data. Furthermore, our framework surpasses pre-trained CNN models in terms of predictive power, affirming its effectiveness in enhancing the accuracy of vibration identification in the presence of constrained SOP data.

Full Article | PDF Article

More Like This

Lifelong QoT prediction: an adaptation to real-world optical networks

Qihang Wang, Zhuojun Cai, and Faisal Nadeem Khan
J. Opt. Commun. Netw. 16(11) 1159-1169 (2024)

Model and data-centric machine learning algorithms to address data scarcity for failure identification

Lareb Zar Khan, João Pedro, Nelson Costa, Andrea Sgambelluri, Antonio Napoli, and Nicola Sambo
J. Opt. Commun. Netw. 16(3) 369-381 (2024)

Digital-twin-assisted meta learning for soft-failure localization in ROADM-based optical networks

Ruikun Wang, Jiawei Zhang, Zhiqun Gu, Memedhe Ibrahimi, Bojun Zhang, Francesco Musumeci, Massimo Tornatore, and Yuefeng Ji
J. Opt. Commun. Netw. 16(7) C11-C19 (2024)

Previous Article Next Article

Cited By

You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Figures (15)

You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Tables (11)

You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Equations (19)

You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Type/Stride	Filter Shape	Input Size
Conv/s2	$3 \times 3 \times 3 \times 32$	$224 \times 224 \times 3$
Conv dw/s1	$3 \times 3 \times 32 d w$	$112 \times 112 \times 32$
Conv/s1	$1 \times 1 \times 32 \times 64$	$112 \times 112 \times 32$
Conv dw/s2	$3 \times 3 \times 64 d w$	$112 \times 112 \times 64$
Conv/s1	$1 \times 1 \times 64 \times 128$	$56 \times 56 \times 64$
Conv dw/s1	$3 \times 3 \times 128 d w$	$56 \times 56 \times 128$
Conv/s1	$1 \times 1 \times 128 \times 128$	$56 \times 56 \times 128$
Conv dw/s2	$3 \times 3 \times 128 d w$	$56 \times 56 \times 128$
Conv/s1	$1 \times 1 \times 128 \times 256$	$28 \times 28 \times 128$
Conv dw/s1	$3 \times 3 \times 256 d w$	$28 \times 28 \times 256$
Conv/s1	$1 \times 1 \times 256 \times 256$	$28 \times 28 \times 256$
Conv dw/s2	$3 \times 3 \times 256 d w$	$28 \times 28 \times 256$
Conv/s1	$1 \times 1 \times 256 \times 512$	$14 \times 14 \times 256$
$5 \times$ conv dw/s1	$3 \times 3 \times 512 d w$	$14 \times 14 \times 512$
$5 \times$ conv/s1	$1 \times 1 \times 256 \times 512$	$14 \times 14 \times 256$
Conv dw/s2	$3 \times 3 \times 512 d w$	$14 \times 14 \times 512$
Conv/s1	$1 \times 1 \times 512 \times 1024$	$7 \times 7 \times 512$
Conv dw/s2	$3 \times 3 \times 1024 d w$	$7 \times 7 \times 1024$
Conv/s1	$1 \times 1 \times 1024 \times 1024$	$7 \times 7 \times 1024$
Avg pool/s1	Pool $7 \times 7$	$7 \times 7 \times 1024$
FC/s1	$1024 \times 1000$	$1 \times 1 \times 1024$
Softmax/s1	Classifier	$1 \times 1 \times 1000$

Approach	Acc (%)	Pre (%)	Re (%)	F1 (%)	Er (%)
Softmax approach (RGB method)	98.5	98.6	98.5	98.5	1.5
Softmax approach (appending method)	97.75	97.8	97.7	97.7	2.25
SVM approach (RGB method)	99.4	99.5	99.4	99.4	0.55
SVM approach (appending method)	98.3	98.4	98.3	98.3	1.66
FF approach	99.6	99.7	99.68	99.68	0.3

Approach	Acc (%)	Pre (%)	Re (%)	F1 (%)	Er (%)
Softmax approach (RGB method)	97	97.27	97	97	3
Softmax approach (appending method)	96.49	97.1	96.4	96.4	3.5
SVM approach (RGB method)	98.6	98.7	98.6	98.7	1.38
SVM approach (appending method)	98.3	98.4	98.3	98.3	1.6
FF approach	99.3	99.4	99.3	99.3	0.62

Appending Order	Acc (%)	F1 (%)	Er (%)
$S_{1} + S_{2} + S_{3}$	98.3	98.3	1.66
$S_{2} + S_{1} + S_{3}$	98.6	98.6	1.38
$S_{3} + S_{1} + S_{2}$	98.6	98.6	1.38
$S_{3} + S_{2} + S_{1}$	98.6	98.6	1.38
$S_{1} + S_{3} + S_{2}$	98	98	1.94
$S_{2} + S_{3} + S_{1}$	98.3	98.3	1.66

Approach	Acc (%)	Pre (%)	Re (%)	F1 (%)	Er (%)
Softmax (RGB method)	98.5	98.6	98.5	98.5	1.5
Softmax (grayscale channel method)	98.2	98.4	98.2	98.2	1.8
SVM (RGB method)	99.4	99.5	99.4	99.4	0.55
SVM (grayscale channel method)	99.1	99.2	99.1	99.1	0.83

Approach	Average Training Time (s)	Average Inference Time per Sample (s)
Softmax approach	23.7	0.002
SVM approach	0.8	0.0007
FF approach	37.97	0.004

Input Size		SVM Approach	FF Approach
$32 \times 32$	Acc	53.6	95.9
	F1	53.6	95.7
	Er	46.3	4.06
$64 \times 64$	Acc	95.83	97.16
	F1	95.8	96.7
	Er	4.16	2.8
$96 \times 96$	Acc	97.7	99.3
	F1	97.7	99.3
	Er	2.22	0.6
$192 \times 192$	Acc	98.9	99
	F1	98.9	99
	Er	1.11	0.93
$224 \times 224$	Acc	99.4	99.6
	F1	99.4	99.6
	Er	0.55	0.3
$240 \times 240$	Acc	99.16	99
	F1	99.1	99
	Er	0.83	0.93
$260 \times 260$	Acc	98.6	98.1
	F1	98.6	98.1
	Er	1.38	1.87

Method	Acc (%)	F1 (%)	Er (%)
Decision tree	87	86.8	13
Naive Bayes	89	89	11
Logistic regression	83	83	17
KNN	89	89	11
SVM approach	99.4	99.4	0.55
FF approach	99.6	99.68	0.3

Pre-trained Model	Acc (%)	F1 (%)	Er (%)	$t_{i n f}$ (s)
VGG16	98	98	1.94	0.001
ResNet50	99.1	99.1	0.83	0.002
Inception	98.8	98.8	1.11	0.002
DenseNet	98	98	1.94	0.0017
MobileNet	99.4	99.4	0.55	0.0007

Data		Softmax	SVM	FF
Unseen dataset	Acc	96	99.1	99.3
Unseen dataset	Er	3.96	0.86	0.68
Synthetic data	Acc	94	97	98
Synthetic data	Er	6	3	2
Noisy data	Acc	94.1	98.8	98.4
Noisy data	Er	6.37	1.2	1.5

Risky event classification leveraging transfer learning for very limited datasets in optical networks

Author Affiliations

ORCID

Abstract

Cited By

Figures (15)

Tables (11)

Equations (19)

Journal of Optical Communications and Networking